GMP with small numbers

[Décio Luiz Gazzoni Filho]

On Jun 18, 2006, at 8:48 AM, Degski wrote:

> On 6/18/06, Paulo J. Matos <pocmatos at gmail.com> wrote:
>>
>> Yes, It seems you're right. Well, less work for me... just need to
>> learn GMP, then! :-)
>>
>> By the way, since I'm programming in C++ I'm considering to use GMP
>> C++ interface. Is there any kind of efficiency issue by using the
>> interface instead of the C functions directly, i.e., is it known  
>> to be
>> slower?
>>
> Well, since the C++ inteface is a wrapper around the C source, it's
> hard to imagine it would get any faster, and it's equally hard to
> imagine that you're not losing any speed in any part of the code, but
> I'm just guessing here.
>
> another clue of course is that GMP is written 99.9% in C. I don't know
> that much about C++, so I wouldn't like to comment on all the
> additional contructs available in C++. But f.e. overloading, which
> although very satisfying from a development point of view (in C you
> would write several different functions to cater for the different
> interfaces), could (not saying allways) lead to unnecessary function
> overhead, in other cases it probably makes no difference whatsoever,
> since in C you'll also need to decide upon (i.e. insert code), which
> funtion to use.

The C++ code uses expression templates, so that overloaded operators  
are resolved to GMP functions at compile time with no runtime  
overhead. Performance is indistinguishable from the C interface's  
performance.

Anyway, the point of expression templates is to avoid unnecessary  
temporaries -- function calls will happen anyway, whether to the  
native C GMP calls or the overloaded operators, it's a function call  
all the same. The only possible worry is in regard to unnecessary  
temporary creation, and expression templates take care of that. You  
might have a point if you use derived classes and virtual function  
calls, but I don't think anyone is going to derive classes from GMP.

To the OP: you can use the C++ without any fear of performance loss.  
Don't be afraid of C++ naysayers (in general, not only those that  
spread misinformation about GMP's interface), they've long been wrong  
about the performance of C++.

Décio
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
Url : http://gmplib.org/list-archives/gmp-discuss/attachments/20060618/d25d81ce/PGP.bin